academic-deep-research

Warn

Audited by Gen Agent Trust Hub on Feb 16, 2026

Risk Level: MEDIUMPROMPT_INJECTIONNO_CODE
Full Analysis
  • Metadata Poisoning (MEDIUM): The README.md claims the skill 'Works offline' and has 'No external dependencies,' yet the research protocol explicitly mandates multiple web_search and web_fetch operations. This contradiction is misleading regarding the skill's data privacy and connectivity requirements.
  • Indirect Prompt Injection (LOW): The skill is designed to ingest and process a high volume of external, untrusted web content via search tools.
  • Ingestion points: Web search results and primary source page content.
  • Boundary markers: Absent. The instructions do not provide delimiters or warnings for the agent to ignore instructions embedded in the retrieved text.
  • Capability inventory: The agent generates narrative reports. It lacks file-writing or system-command capabilities in the provided documentation.
  • Sanitization: Absent. The skill relies on the agent's reasoning without technical filtering of the input.
  • Missing Component (LOW): The README references a SKILL.md file containing '500+ lines' of methodology which was not provided for analysis. This prevents a full verification of the primary logic.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 16, 2026, 05:00 AM